NBit forward TBE: remap indices before prefetching; and some more unit tests. #1433
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
✅ Deploy Preview for pytorch-fbgemm-docs canceled.
|
|
This pull request was exported from Phabricator. Differential Revision: D40821768 |
doehyun
added a commit
to doehyun/FBGEMM
that referenced
this pull request
Oct 31, 2022
…t tests. (pytorch#1433) Summary: Pull Request resolved: pytorch#1433 It's reported that UVM_CACHING doesn't work with pruning. D40788589 (pytorch@bcc69ed) fixed issues in linearize kernel. It worked with int_nbit_split_embedding_uvm_caching_codegen_lookup_function, because in that case, index remapping for pruning occurs before calling int_nbit_split_embedding_uvm_caching_codegen_lookup_function. IntNBitTableBatchedEmbeddingBagsCodegen::forward() runs index remapping after prefetch, so it didn't work even with the fix in linearize kernel. This diff changes the order we call index remapping: first, index remapping, and then prefetch. Added related tests also. Differential Revision: D40821768 fbshipit-source-id: 3ee17e95ee63b3ba112f67824f0b586749e7b8e7
doehyun
force-pushed
the
export-D40821768
branch
from
817c0f2
to
ea92d8a
Compare
|
This pull request was exported from Phabricator. Differential Revision: D40821768 |
doehyun
added a commit
to doehyun/FBGEMM
that referenced
this pull request
Nov 4, 2022
…t tests. (pytorch#1433) Summary: Pull Request resolved: pytorch#1433 It's reported that UVM_CACHING doesn't work with pruning. D40788589 (pytorch@bcc69ed) fixed issues in linearize kernel. It worked with int_nbit_split_embedding_uvm_caching_codegen_lookup_function, because in that case, index remapping for pruning occurs before calling int_nbit_split_embedding_uvm_caching_codegen_lookup_function. IntNBitTableBatchedEmbeddingBagsCodegen::forward() runs index remapping after prefetch, so it didn't work even with the fix in linearize kernel. This diff changes the order we call index remapping: first, index remapping, and then prefetch. Added related tests also. Differential Revision: D40821768 fbshipit-source-id: 5dca0948f4714864c2d0bb94a360cfe659e9f8f6
doehyun
force-pushed
the
export-D40821768
branch
from
ea92d8a
to
814c3a7
Compare
|
This pull request was exported from Phabricator. Differential Revision: D40821768 |
1 similar comment
|
This pull request was exported from Phabricator. Differential Revision: D40821768 |
doehyun
added a commit
to doehyun/FBGEMM
that referenced
this pull request
Nov 18, 2022
…t tests. (pytorch#1433) Summary: Pull Request resolved: pytorch#1433 It's reported that UVM_CACHING doesn't work with pruning. D40788589 (pytorch@bcc69ed) fixed issues in linearize kernel. It worked with int_nbit_split_embedding_uvm_caching_codegen_lookup_function, because in that case, index remapping for pruning occurs before calling int_nbit_split_embedding_uvm_caching_codegen_lookup_function. IntNBitTableBatchedEmbeddingBagsCodegen::forward() runs index remapping after prefetch, so it didn't work even with the fix in linearize kernel. This diff changes the order we call index remapping: first, index remapping, and then prefetch. Added related tests also. Differential Revision: D40821768 fbshipit-source-id: db8f466d65722dcbb8ecab44da3abc28ae3d476d
doehyun
force-pushed
the
export-D40821768
branch
from
814c3a7
to
18377a2
Compare
|
This pull request was exported from Phabricator. Differential Revision: D40821768 |
doehyun
added a commit
to doehyun/FBGEMM
that referenced
this pull request
Dec 8, 2022
…t tests. (pytorch#1433) Summary: Pull Request resolved: pytorch#1433 It's reported that UVM_CACHING doesn't work with pruning. D40788589 (pytorch@bcc69ed) fixed issues in linearize kernel. It worked with int_nbit_split_embedding_uvm_caching_codegen_lookup_function, because in that case, index remapping for pruning occurs before calling int_nbit_split_embedding_uvm_caching_codegen_lookup_function. IntNBitTableBatchedEmbeddingBagsCodegen::forward() runs index remapping after prefetch, so it didn't work even with the fix in linearize kernel. This diff changes the order we call index remapping: first, index remapping, and then prefetch. Added related tests also. Differential Revision: D40821768 fbshipit-source-id: 6dedc13e4fe9c006ee00df833837d224b43a9257
doehyun
force-pushed
the
export-D40821768
branch
from
18377a2
to
ffb7903
Compare
doehyun
added a commit
to doehyun/FBGEMM
that referenced
this pull request
Dec 8, 2022
…t tests. (pytorch#1433) Summary: Pull Request resolved: pytorch#1433 It's reported that UVM_CACHING doesn't work with pruning. D40788589 (pytorch@bcc69ed) fixed issues in linearize kernel. It worked with int_nbit_split_embedding_uvm_caching_codegen_lookup_function, because in that case, index remapping for pruning occurs before calling int_nbit_split_embedding_uvm_caching_codegen_lookup_function. IntNBitTableBatchedEmbeddingBagsCodegen::forward() runs index remapping after prefetch, so it didn't work even with the fix in linearize kernel. This diff changes the order we call index remapping: first, index remapping, and then prefetch. Added related tests also. Differential Revision: D40821768 fbshipit-source-id: 017db632cd2d7891c177cd3959f1d8967538b2ea
doehyun
force-pushed
the
export-D40821768
branch
from
ffb7903
to
2d6f38c
Compare
|
This pull request was exported from Phabricator. Differential Revision: D40821768 |
…t tests. (pytorch#1433) Summary: Pull Request resolved: pytorch#1433 It's reported that UVM_CACHING doesn't work with pruning. D40788589 (pytorch@bcc69ed) fixed issues in linearize kernel. It worked with int_nbit_split_embedding_uvm_caching_codegen_lookup_function, because in that case, index remapping for pruning occurs before calling int_nbit_split_embedding_uvm_caching_codegen_lookup_function. IntNBitTableBatchedEmbeddingBagsCodegen::forward() runs index remapping after prefetch, so it didn't work even with the fix in linearize kernel. This diff changes the order we call index remapping: first, index remapping, and then prefetch. Added related tests also. Reviewed By: jspark1105 Differential Revision: D40821768 fbshipit-source-id: 03487fe137a76b1d4819aa8c6070bc5c49ac60a8
doehyun
force-pushed
the
export-D40821768
branch
from
2d6f38c
to
095ab3b
Compare
|
This pull request was exported from Phabricator. Differential Revision: D40821768 |
|
This pull request has been merged in 2cb8b0d. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary:
It's reported that UVM_CACHING doesn't work with pruning.
D40788589 (bcc69ed) fixed issues in linearize kernel.
It worked with int_nbit_split_embedding_uvm_caching_codegen_lookup_function, because
in that case, index remapping for pruning occurs before calling int_nbit_split_embedding_uvm_caching_codegen_lookup_function.
IntNBitTableBatchedEmbeddingBagsCodegen::forward() runs index remapping after
prefetch, so it didn't work even with the fix in linearize kernel.
This diff changes the order we call index remapping: first, index remapping, and then prefetch.
Added related tests also.
Differential Revision: D40821768